torphix / Music-Genre-Classification

Traditional machine learning techniques + Neural networks for music genre classification

Setup

• Note: code has been tested on linux and macOS we have tried to keep the environment as OS agnostic as possible however as we don't own a windows machine we cannot on windowsOS.
• Due to package conflicts between tensorflow and streamlit (package used to create UI) it is necessary to use a seperate environment for viewing the UI

Keras Neural Network Environment Setup

conda env create -f conda.yaml
conda activate music
export XLA_FLAGS=--xla_gpu_cuda_data_dir=/usr/lib/cuda
pip install -r requirements/requirements_keras.txt
In order to configure different parameters such as input datatypes, model sizes etc look under configs/keras_config.yaml

UI Environment Setup

Only required for running the UI and legacy torch code

conda deactivate
python -m venv venv
source venv/bin/activate
pip install -r requirements/requirements_ui.txt

Commands

• run command python ui.py (Activate UI environment first) to launch the UI providing a succinct overview of the project.

• run command python main.py process_data (Activate Keras Neural Network setup first)

• run command python main.py train_nn_keras (Activate Keras Neural Network setup first) must run python main.py process_data first

• run command python main.py fit_classical_models to run the classical machine learning models (Activate Keras Neural Network setup first)

• run command python main.py evaluate_classical_models to evaluate the classical machine learning models (Activate Keras Neural Network setup first)

• open notebook Classical ML data analysis.ipynb for a brief analysis with classical ml models.